[llvm] [X86] `combinePMULH` - combine `mulhu` + `srl` (PR #132548)

[Simon Pilgrim via llvm-commits]

================
@@ -54231,10 +54231,16 @@ static SDValue combinePMULH(SDValue Src, EVT VT, const SDLoc &DL,
 
   // First instruction should be a right shift by 16 of a multiply.
   SDValue LHS, RHS;
+  APInt ShiftAmt;
   if (!sd_match(Src,
-                m_Srl(m_Mul(m_Value(LHS), m_Value(RHS)), m_SpecificInt(16))))
+                m_Srl(m_Mul(m_Value(LHS), m_Value(RHS)), m_ConstInt(ShiftAmt))))
     return SDValue();
 
+  if (ShiftAmt.ult(16))
----------------
RKSimon wrote:

`if (ShiftAmt.ult(16) || ShiftAmt.uge(InVT.getScalarSizeInBits()))` - because eventually a fuzz test will come along and try to break things :( 

https://github.com/llvm/llvm-project/pull/132548